Theory:

Natural Language Processing: Textbook by Dan Jurafsky and James H. Martin 

Stanford CS2N: Natural Language Processing with Deep Learning: Stanford CS224N 

Machine Learning and Big Data—Melissa Dell and Matthew Harding: https://www.aeaweb.org/conference/cont-ed/2023-webcasts

A brief introduction to NLP of Chinese text data.: https://carlos9310.github.io/assets/pdf/chinese-nlp.pdf  

OCR: 

ABBYY performs well in OCRing Chinese PDF files such as modern gazetteers. 

Baiduyun can also handle large-scale OCR if you subscribe to its SVIP membership, which allows you to submit five files (smaller than 200 mb) at a time with no limit on the number of batches.

Cons: It may censor the content. 

Pros: You can split large files into segments smaller than 200 MB each, submit them to Baidu Cloud for OCR, and then merge the results back into a single document.

The Qwen-VL OCR model is highly cost-effective. Processing a file with 600 to 1,000 pages costs approximately 1 RMB.

Shen et al. (2021) OCR Tables and Parse the Output 

Pros: OCR software frequently misidentifies 4/A, 8/B, 0/O, and other characters. The program created by Prof. Dell could address these issues.

Cons: My experience with the PDF version of the Japanese census shows that the programming could extract half of the data without needing to double-check it manually.

Academia Sinica Text Recognition and Proofreading Platform: https://ocr.ascdc.tw/ is perfect in OCRing traditional Chinese in vertical columns. 

Empirical:

Text Similarity: Text similarity using text2vec 

Supervised Learning: https://www.youtube.com/watch?v=BAP6l2uGAHU  A practical example of Chinese text data classification using RNN.

Unsupervised Learning: https://github.com/huang027/LDA  Latent Dirichlet Allocation Topic Model

Toolkit: https://github.com/jiaeyan/Jiayan#2  Jiayan, the 1st NLP toolkit designed for Classical Chinese, supports lexicon construction, tokenizing, POS tagging, sentence segmentation, and punctuation.